[llvm] LangRef: storing poison in memory is UB (PR #141339)

[Nuno Lopes via llvm-commits]

nunoplopes wrote:

I don't think it's that hard to fix. Widening happens just in a few places. (like lowering on memory intrinsics such as memcpy/memcmp) and some codegen prepare.

The issue is that stores of poison can happen anywhere. And as I mentioned, making it UB probably makes a bunch of optimizations wrong. We would be trading of a few wrong optimizations with another few wrong optimizations 😅

https://github.com/llvm/llvm-project/pull/141339